void operator()(Type* dest, const math::Size_t<1> pitchDest, const Type* source, const math::Size_t<1> pitchSource, const math::Size_t<2u>& size, flags::Memcopy::Direction direction) { const cudaMemcpyKind kind[] = {cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, cudaMemcpyHostToHost, cudaMemcpyDeviceToDevice}; CUDA_CHECK_NO_EXCEP(cudaMemcpy2D(dest, pitchDest.x(), source, pitchSource.x(), sizeof(Type) * size.x(), size.y(), kind[direction])); }
void operator()(Type* dest, const math::Size_t<2u> pitchDest, Type* source, const math::Size_t<2u> pitchSource, const math::Size_t<3>& size, flags::Memcopy::Direction direction) { const cudaMemcpyKind kind[] = {cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, cudaMemcpyHostToHost, cudaMemcpyDeviceToDevice}; cudaPitchedPtr pitchedPtrDest; pitchedPtrDest.pitch = pitchDest.x(); pitchedPtrDest.ptr = dest; pitchedPtrDest.xsize = size.x(); pitchedPtrDest.ysize = size.y(); cudaPitchedPtr pitchedPtrSource; pitchedPtrSource.pitch = pitchSource.x(); pitchedPtrSource.ptr = source; pitchedPtrSource.xsize = size.x(); pitchedPtrSource.ysize = size.y(); cudaMemcpy3DParms params; params.srcArray = NULL; params.srcPos = make_cudaPos(0,0,0); params.srcPtr = pitchedPtrSource; params.dstArray = NULL; params.dstPos = make_cudaPos(0,0,0); params.dstPtr = pitchedPtrDest; params.extent = make_cudaExtent(size.x() * sizeof(Type), size.y(), size.z()); params.kind = kind[direction]; CUDA_CHECK_NO_EXCEP(cudaMemcpy3D(¶ms)); }
/** * Creates a host buffer from a pointer with a size. Assumes dense layout (no padding) * * @param ptr Pointer to the first element * @param size Size of the buffer * @param ownMemory Set to false if the memory is only a reference and managed outside of this class * @param pitch Pitch in bytes (number of bytes in the lower dimensions) */ HINLINE HostBuffer(Type* ptr, const math::Size_t<T_dim>& size, bool ownMemory, PitchType pitch = PitchType::create(0)) { this->dataPointer = ptr; this->_size = size; if(T_dim >= 2) this->pitch[0] = (pitch[0]) ? pitch[0] : size.x() * sizeof(Type); if(T_dim == 3) this->pitch[1] = (pitch[1]) ? pitch[1] : this->pitch[0] * size.y(); this->refCount = new int; *this->refCount = (ownMemory) ? 1 : 2; }
void operator()(Type* dest, const math::Size_t<0>, const Type* source, const math::Size_t<0>, const math::Size_t<1>& size, flags::Memcopy::Direction direction) { const cudaMemcpyKind kind[] = {cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, cudaMemcpyHostToHost, cudaMemcpyDeviceToDevice}; CUDA_CHECK(cudaMemcpy(dest, source, sizeof(Type) * size.x(), kind[direction])); }
dim3 cudaGridDim(const math::Size_t<2>& size) const { return dim3(size.x() / BlockDim::x::value, size.y() / BlockDim::y::value, 1); }
dim3 cudaGridDim(const math::Size_t<3>& size) const { return dim3((size.x() / BlockDim::x::value) * (size.z() / BlockDim::z::value), size.y() / BlockDim::y::value, 1); }
SphericMapper(const math::Size_t<3>& size) : widthInBlocks(size.x() / BlockDim::x::value) {}
PNGBuffer(math::Size_t<2> size, const std::string& name) : png(size.x(), size.y(), 0.0, name.data()), size(size) {}